Inherent AI’s Faraday Beats Larger Models in Scientific Paper Replication

Key Points

• Inherent’s Faraday outperformed Anthropic’s Claude Opus 4.8 and OpenAI’s GPT‑5.5 on a paper‑replication benchmark.
• Faraday runs on Qwen 3.6, a 27‑billion‑parameter model, far smaller than the competing systems.
• The agent was trained with reinforcement learning to develop a sense of research "taste".
• Inherent leverages OpenAI’s GPT‑5.5 Codex for coding rather than building its own tool.
• The London startup raised a $50 million seed round and aims to grow to 20‑25 staff by year‑end.
• Co‑founder Edward Hughes criticized the U.K.’s garden‑leave rules that delay talent movement.